AI Model Revolutionizes Embryo Selection in IVF Procedures
An innovative AI-powered model developed by researchers at Weill Cornell Medicine promises to transform the embryo selection process in in vitro fertilization (IVF), according to [NVIDIA](https://developer.nvidia.com/blog/time-lapse-ai-model-enhances-ivf-embryo-selection/). The model, named the Blastocyst Evaluation Learning Algorithm (BELA), leverages time-lapse imaging data and maternal age to assess embryo quality and chromosomal health.
Advancements in IVF Technology
IVF, a method that has enabled over eight million births since 1978, relies heavily on accurate embryo selection to improve pregnancy success rates. Traditional methods, such as preimplantation genetic testing for aneuploidy (PGT-A), involve cell extraction, which can be costly and potentially harmful to embryo viability. BELA presents a non-invasive and cost-effective alternative, potentially streamlining the selection process and lowering expenses for families.
How BELA Works
BELA automates the evaluation of embryos by analyzing time-lapse imaging data over five days of development. This data, combined with maternal age, allows the AI model to predict chromosomal health and rank embryos by quality. The model’s accuracy hinges on its ability to assess timing and speed as indicators of embryo viability.
Developed on Cornell’s BioHPC computing cluster with NVIDIA A40 GPUs, BELA processes data efficiently, requiring just 5.23 minutes for training and approximately 30 seconds per embryo prediction. The AI model was trained on a diverse dataset of over 2,800 embryo sequences, enhancing its reliability and precision.
Clinical Application and Performance
To facilitate clinical use, the research team also created STORK-V, a web-based platform powered by BELA. This tool allows embryologists to upload time-lapse imaging data and receive real-time predictions on embryo quality and chromosomal health. BELA has demonstrated superior performance compared to existing AI models, achieving an AUC of 0.82 in distinguishing normal from abnormal embryos. Its predictions rival the accuracy of manual evaluations by embryologists.
Implications for IVF Practices
While BELA is not intended to replace PGT-A, it serves as a valuable prescreening tool, aiding embryologists in deciding which embryos warrant further analysis. This approach can reduce costs and enhance the efficiency of IVF procedures, ensuring that viable embryos are prioritized for implantation.
The open-source code for BELA is accessible on GitHub, allowing further development and integration into clinical settings. The study detailing BELA's capabilities is published in Nature Communications.